|
|
| Nonrandomized Comparative Clinical Studies - Proceedings of
the International Conference on Nonrandomized Comparative
Clinical Studies in Heidelberg, April 10 -11,1997 |
 |
|
Modelling unobserved non-compliance in clinical studies
E. Dietz, D. Boehning
Abstract
 Not considering unobserved
non-compliance in clinical studies can lead to seriously biased effect estimates. A simple
mixture model is proposed to take non-differential and also differential compliance into
consideration, where compliance is called differential if its probability depends on the
response variable. It is a two-component mixture, where the components are the model of
the compliers and the model of the non-compliers, repectively. The respective mixture
weights may depend on the dose of the active substance, covariables, and the response
variable. In this paper, procedures are given, which provide maximum likelihood estimators
of the unknown model parameters as well as their standard errors. These are demonstrated
by means of example data. Other tools for statistical inference like measurements of local
and global goodness of fit are proposed. Problems of causal inference when using this
model are discussed.
References
- [1]
- Efron, B., Feldman, D. (1991) Compliance as an explanatory variable in
clinical trials. Journal of the American Statistical Association 86, 9-26.
- [2]
- Dietz, E., Boehning, D.(1995) Statistical inference based on a general
model of unobserved heterogeneity. Lecture Notes in Statistics 104, 75-82.
- [3]
- Dietz, E., Boehning, D.(1996) The use of two-component mixture models
with one completely or partly known component. Computational statistics to appear.
|
